Abstract
The Icelandic population is scarce compared to the Norwegian, causing high variability in the observed mortalities. We apply two mortality models to data on the Norwegian population dating 60 years back, and the Icelandic 30 years back. A modified smooth version of the Lee-Carter model and a new model introduced by Erik Bølviken. The analysis is done using a Bayesian approach performing inference via Markov Chain Monte Carlo (MCMC), where the aim is to estimate smooth mortality curves resem- bling the unknown underlying mortality and, further on, use the models to forecast future mortality. We use different statistical methods to determine model comparison and goodness of fit, such as the Deviance Information Criterion (DIC) and posterior predictive p-values. To judge the forecasts we use the verification method Continuous Ranked Probability Score (CRPS).
The Icelandic population is scarce compared to the Norwegian, causing high variability in the observed mortalities. We apply two mortality models to data on the Norwegian population dating 60 years back, and the Icelandic 30 years back. A modified smooth version of the Lee-Carter model and a new model introduced by Erik Bølviken. The analysis is done using a Bayesian approach performing inference via Markov Chain Monte Carlo (MCMC), where the aim is to estimate smooth mortality curves resem- bling the unknown underlying mortality and, further on, use the models to forecast future mortality. We use different statistical methods to determine model comparison and goodness of fit, such as the Deviance Information Criterion (DIC) and posterior predictive p-values. To judge the forecasts we use the verification method Continuous Ranked Probability Score (CRPS).